How AI Helped Me Finish A Civil Engineering Project Two Weeks Early, And Where It Nearly Slowed Me Down Instead
A real account of using AI tools across a single site project from initial review to final submission, including the point where trusting the output too quickly almost created a costly mistake.
Marcus Webb
July 9, 2026
Project: a small commercial building fit out with a compressed timeline and a client who requested weekly progress updates. Original schedule: 10 weeks. Actual completion: 8 weeks. Where AI nearly caused a problem: a compliance summary that omitted a regional fire code update because the source document it referenced was slightly out of date, caught during a manual final check before submission.
Why This Project Was A Real Test, Not A Showcase
I did not set out to write a case study, the deadline was simply tight enough that I needed every tool available to work, not just sound impressive in a demo. This is the honest sequence of what happened, including the part I almost got wrong.
Week One: Drawing Review Cut From Two Days To Half A Day
Running the initial plan set through Civils.ai flagged three clearance issues on the first pass that would normally take a full manual review to catch. It also flagged one issue that turned out to be a false positive, a fixture placement that was actually compliant under a regional exception the tool did not account for, so I still needed to verify every flag rather than act on it blindly.
Weeks Two Through Five: Weekly Client Updates Without Losing Evenings To Writing Them
- OpenSpace captured weekly site walkthroughs automatically, giving a visual record without extra time spent documenting manually
- Claude turned my rough bullet point progress notes into a clean client facing weekly update in under ten minutes each time
- This alone likely saved 3 to 4 hours per week that used to go into formatting and rewriting the same update from scratch
Week Six: The Mistake That Almost Happened
I asked Claude to summarize a compliance document against the current fire code for the region. The summary read as complete and confident, which is exactly the problem, because the source document I had provided was an older version I had not realized was superseded three months earlier. A colleague doing an unrelated review caught the discrepancy by checking the regulator's site directly. Nothing shipped incorrectly, but it was close, and it reinforced a rule I now follow strictly: never trust an AI summary of a regulatory document without verifying the source document itself is current.
Weeks Seven And Eight: Finishing Early
With the compliance verification tightened up and the review and reporting workflow running smoothly, the final two weeks moved faster than the original schedule accounted for. The project closed two weeks ahead of the original 10 week estimate, with the time saved coming almost entirely from documentation and review speed rather than from any change to the actual construction or design work itself.
What I Changed In My Process After This Project
- Always confirm the source document version date before asking any AI tool to summarize a regulatory or code document
- Treat AI flagged issues on drawing review as a starting checklist, not a final answer
- Keep AI generated client communications in my own voice by editing the draft rather than sending it unedited
- Build in one manual final check on anything compliance related before submission, no exceptions, regardless of how confident the AI summary sounds
Final Thoughts
The time savings were real and the project genuinely finished early because of these tools, but the near miss on the compliance summary is the part I think about more. AI tools in this field are excellent at speeding up review and documentation and completely unreliable as a final source of truth on anything with regulatory or safety weight. Treating them as a fast first pass rather than a final answer is what made this project work, and it is the only way I would recommend using them going forward.